Abstract
The risk-neutral density represents the probability distribution of future asset prices (or returns) under the risk-neutral measure. It is a pivotal probability distribution, encompassing crucial insights into asset-price dynamics and investor preferences, thus holding a central place in theoretical and empirical finance. In this thesis, I delve into the intricate domain of risk-neutral density, employing generative networks to introduce innovative methodologies that address complex challenges in this field. Through two pivotal essays, the thesis explores diverse yet interconnected problems, highlighting the adaptability and profound potential of generative machine learning techniques in advancing our understanding of risk-neutral density and tail risk premiums.This thesis explores the intersection of generative machine learning and financial risk management through two essays focused on risk-neutral density (RND) estimation and its applications. By addressing key challenges in modern finance, it offers innovative methodologies that contribute both theoretical advancements and practical tools for market analysis.
The first essay introduces the Risk-Neutral Generative Network (RNGN), a generative machine-learning-based framework for extracting RNDs from option prices. This approach combines the flexibility of generative models with financial interpretability, ensuring compliance with no-arbitrage principles. Three model variations are developed to cater to different levels of complexity and data characteristics, enabling accurate estimation of RNDs across diverse market scenarios. Extensive numerical studies, including simulations and empirical tests on S&P 500 options, highlight the RNGN’s superiority over classical models in capturing intricate distributional features, such as skewness and kurtosis, while maintaining robustness under various market conditions.
The second essay extends the utility of the RNGN framework to investigate tail risk premiums, a critical component of financial risk management. By leveraging the flexibility of the proposed models, it constructs tail risk measures that exhibit strong predictive power for market returns, particularly in short horizons, and demonstrates pricing capabilities in cross-sectional asset analysis. The study also provides evidence of the tail risk premium’s significance in forecasting extreme market events, underscoring its practical value for risk managers.
This thesis contributes to the growing synergy between machine learning and financial theory. It demonstrates how advanced generative models can address challenges in RND estimation and tail risk analysis, offering a path to more effective risk management and asset pricing strategies. Beyond its immediate applications, this research opens avenues for future exploration, such as integrating real-world constraints, applying the framework to other asset classes, and adopting alternative machine learning architectures. By bridging methodological innovation with real-world relevance, the work establishes a foundation for harnessing the transformative potential of artificial intelligence in finance.
| Date of Award | 24 Apr 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Qi WU (Supervisor) |